Vol.:(0123456789) 1 3
Health and Technology
https://doi.org/10.1007/s12553-020-00506-6
ORIGINAL PAPER
Automatic suspicions lesions segmentation based on variable‑size
windows in mammography images
Bahram Sadeghi
1
· Meysam Karimi
1
· Samaneh Mazaheri
2
Received: 11 May 2020 / Accepted: 29 October 2020
© IUPESM and Springer-Verlag GmbH Germany, part of Springer Nature 2020
ABSTRACT
Breast cancer is the second main cause of death in women of western countries, so early detection and prevention is crucial.
Early detection increases the likelihood of treatment as well as patient resistance. Among breast cancer detection methods,
mammography is the most efective diagnostic method. For radiologists, diagnosing a cancerous mass on mammographic
images is prone to error, which shows there is a need for a method to reduce the errors. In this study, a new adaptive threshold-
ing method is proposed based on variable-sized windows. This method estimates the location of the mass and then determines
the exact location of the cancerous tissue to reduce false positives. To detect the mass automatically, frstly, the histograms
diagram and its relative maximums have been used to calculate the initial threshold for estimating the mass location. Two
windows that contain information around each pixel and their size varies according to the mean value of each image due to
the preservation of useful information. Secondly, two windows are used for the fnal threshold in order to discover the loca-
tion of the mass and its exact shape. The proposed approach has been applied to 170 images of the Mammographic Image
Analysis Society MiniMammographic database. Evaluations have shown 96.7% sensitivity and 0.79 false-positive rates,
which prove an improvement in comparison with other state-of-the-art methods.
Keywords Texture · Mammography · Segmentation · Computer-aided detection systems · Breast cancer · Adaptive
thresholding
1 Introduction
Breast cancer is the most common form of cancer among
women. Tumors of breast cancer are small at frst and some-
times it takes several years to become a large gland, So early
detection of tissue is ciritial. Breast cancer detection meth-
ods include: Breast self-examination, Breast examination by
a doctor and Mammography, which the most efective way to
detect breast cancer is Mammography. Incorrect detections
in mammograpgy images by radiologists errors of analysing
diagnosis due to physician fatigue or optical illusion, lead
to the idea of fnding a solution to do the process automati-
cally. Radiologists fail to detect tissue in 10 to 30 percent of
cases. In order to reduce these human mistakes, computer-
aided detection/diagnosis (CAD) systems have been used
[8]. Computer-aided detection systems are the most efective
tools for early diagnosis of breast cancer. Segmentation is
considered as one of the main steps in image processing.
It divides digital images into multiple regions for better
analysis. It can also specifes diferent objects in the image.
Several segmentation techniques for image smoothing and
easy evaluation have been developed by researchers. The
most popular image segmentation techniques are: Edge
Detection, Thresholding, Histogram, Region based meth-
ods and Watershed Transformation. Images are divided into
two types based on their color (RGB images and Grayscale
images). Therefore, the segmentation of RGB images and
Grayscale images are completely diferent. Specifcations of
the pixels and information about their adjacent pixels are two
fundamental parameters for any segmentation algorithm.
* Samaneh Mazaheri
Samaneh.Mazaheri@ontariotechu.ca
Bahram Sadeghi
b_sadeghi_b@iasbs.ac.ir
Meysam Karimi
meysam.k@iasbs.ac.air
1
Department of Computer Science and Technology, Institute
for Advanced Studies in Basic Sciences (IASBS), Zanjan,
Iran
2
Faculty of Business and Information Technology, Ontario
Tech University, Oshawa, ON, Canada